android-crash-journal-and-periodic-work-verify
Prove an Android app's own crash journal records a real crash and that its periodic WorkManager job is actually registered with the OS — covers the am crash routing, the below-the-fold Settings trap t
SKILL.mdskills/managed/android-crash-journal-and-periodic-work-verify/SKILL.md
GitHubVerifying a crash journal and a periodic job on device
Two claims that are usually only checked at the call site: "crashes are recorded" and "it syncs on a schedule". Both are verifiable in minutes, and both have a trap that makes a working feature look broken.
1. Force a real crash through the app's handler
am crash delivers a RemoteServiceException through the app process, so
a Thread.setDefaultUncaughtExceptionHandler installed at startup does run:
adb shell am crash <pkg>
sleep 4
adb shell pidof <pkg> # empty: the process died, as it should
Then relaunch and read the record off disk (debuggable builds only):
adb shell run-as <pkg> ls -l /data/data/<pkg>/files/crash/
adb shell run-as <pkg> cat /data/data/<pkg>/files/crash/<file>
A record worth having names app version, API level, device, thread and the
stack. If the directory exists but is empty, the handler was installed after
the crash path or was replaced — check it is installed unconditionally in
Application.onCreate, not behind a debug flag.
Clean up afterwards. A synthetic crash left in the journal is a fake defect for whoever reads it next:
adb shell run-as <pkg> rm -f /data/data/<pkg>/files/crash/<file>
2. The trap: the UI section is below the fold
A crash-count row usually sits near the bottom of a settings screen. A single
uiautomator dump right after opening Settings reports no crash text at
all, which reads exactly like "the journal did not record". Scroll before
concluding:
for i in range(10):
x = dump()
if "crash record" in x or "No crashes recorded" in x: break
sh("shell","input","swipe","540","1500","540","700","300")
Same class as tapping stale coordinates: settle the screen, then read.
3. Prove the periodic job exists at the OS
schedule() being called proves nothing — WorkManager may have rejected,
replaced, or never persisted it. Ask the system:
adb shell dumpsys jobscheduler | grep -n -i <pkg>
Look for a job whose Service is the app's own:
JOB androidx.work.systemjobscheduler:uXXX/0
Service: <pkg>/androidx.work.impl.background.systemjob.SystemJobService
Requires: charging=false batteryNotLow=true deviceIdle=false
Unsatisfied constraints: TIMING_DELAY
Identify it as yours by its constraints. batteryNotLow=true and nothing
else matches a request built with
Constraints.Builder().setRequiresBatteryNotLow(true). That correspondence is
what turns "a job exists" into "our job was accepted".
TIMING_DELAY unsatisfied is the healthy state for a periodic job between
runs — not a fault.
4. What this does NOT tell you
The interval is not printed in that dumpsys output. A job scheduled with
the wrong period looks identical. Say so rather than implying it was observed:
source the interval from the code (PeriodicWorkRequestBuilder(Duration.ofDays(1)))
and mark it as unverified-on-device.
Splitting dumpsys jobscheduler on JOB # also mis-attributes blocks to other
packages — grep for the package and read the surrounding lines instead.